adversarial-review
Pass
Audited by Gen Agent Trust Hub on Jul 20, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill methodology requires agents to ingest and process external code and data structures, which establishes a surface for indirect prompt injection where the material being reviewed could influence the agent's behavior.
- Ingestion points: Target source files, schemas, contracts, and protocol definitions as identified in the SKILL.md charter template.
- Boundary markers: The instructions lack specific boundary delimiters or guidelines for the agent to ignore instructions embedded within the target files.
- Capability inventory: The skill directs the agent to execute code by building scratch harnesses and running binaries to perform dynamic analysis.
- Sanitization: There are no provisions for sanitizing or validating the code surfaces or the hostile inputs generated by the subagents.
Audit Metadata